[Eeglablist] AMICA stopping criteria?
elrawls at email.uark.edu
Fri Nov 3 07:25:57 PDT 2017
I am curious about the stopping criteria for amica15. It looks like the
default is for the algorithm to stop at 2000 iterations, or a gradient of
1e-7, or an LL change of 1e-9, whichever comes first.
However, every time I've run amica it is stopped only by reaching the 2000
iter limit; it never reaches the gradient or LL cutoff for stopping. The
gradient generally hovers somewhere around 1e--5 and the LL changes by 1e-6
by the last iteration.
How likely is it for amica to actually hit these default cutoff values?
Does this indicate failure in separation, or is this essentially just a
noise threshold which it is unlikely amica will pass? It is worth noting
that the last time I ran amica on a dataset it hit an nd of 2e-5 by iter
875 and stayed essentially at the same level until iter 2000.
Graduate Research Assistant
University of Arkansas
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the eeglablist